park

name

park

description

Park the current session by generating all session artifacts, archiving them to $SESSION_KIT_ROOT/sessions/<project>/<date-label>/ (default ~/.stoobz/), and updating the manifest. Use when the user says "/park", "park this session", "wrap up", "I'm done for now", "save everything", or wants to create a complete handoff package before leaving a session. Runs /tldr, /relay, and /hone in sequence, then archives. Supports --archive-system for retroactive cleanup of scattered .stoobz/ directories (with --select, --all, --dry-run, --clean flags).

Park Session

End-of-session ceremony. Generate TLDR / RELAY / HONE, then finalize the active archive — rename <sid>-active/ to <date>-<label>/, flip the manifest entry to archived, and append chain metadata to the relay baton.

Three sk write-artifact calls and one sk park-finalize call. The judgment (label resolution, tag inference, summary extraction) stays in this skill's prose; the durability, locking, and manifest math live in the binaries.

Contract

• Inputs: optional <label> argument (e.g., /park scripts-as-tools-substrate); current conversation; current git state.
• Outputs: three artifacts in the durable archive (TLDR.md, CONTEXT_FOR_NEXT_SESSION.md, HONE.md); cwd mirror of the same; manifest entry for this session flipped from active → archived; chain metadata block appended to ./.stoobz/CONTEXT_FOR_NEXT_SESSION.md.
• Idempotent: if the session is already archived, sk park-finalize is a friendly no-op — it surfaces the existing archive path and exits 0.

Process

1. Announce

Tell the user: Parking this session. Generating artifacts...

2. Compose the three artifact bodies

For each of TLDR / RELAY / HONE, follow the body-composition prose in that skill's Process section (sections 1–4 of tldr/SKILL.md, relay/SKILL.md, hone/SKILL.md). Bind the resulting bodies to shell variables $TLDR_BODY, $RELAY_BODY, $HONE_BODY.

Read prior archive copies (~/.stoobz/sessions/<project>/<sid>-active/<NAME>.md) and weave their content under a ## Previous Session heading if present — the rolling-history rule from each individual skill applies here too.

3. Write the three artifacts via the substrate

echo "$TLDR_BODY"  | sk write-artifact --skill park --artifact TLDR.md --content-stdin
echo "$RELAY_BODY" | sk write-artifact --skill park --artifact CONTEXT_FOR_NEXT_SESSION.md --content-stdin
echo "$HONE_BODY"  | sk write-artifact --skill park --artifact HONE.md --content-stdin

--skill park (not tldr / relay / hone) so the ledger attributes the writes to the park ceremony — "what wrote this artifact" stays answerable when /park runs the same artifacts the individual skills also produce.

Each call performs the four-step durable-first protocol: checkin precondition → archive write → verify → ledger append → cwd mirror. See write-artifact-protocol.md.

4. Resolve label (first match wins)

Priority	Source	Use when
1	/park <label> argument	Operator passed an explicit label
2	Git branch name	Branch is not main / master / develop
3	Slugified first heading of just-written TLDR.md	No useful branch name
4	(empty — date-only archive)	All else fails

Slugify rules: lowercase, ASCII, alphanumeric + -, max 50 chars, no leading/trailing -.

5. Resolve summary

Use the title of the just-written TLDR (the text after # TLDR: on the first heading line). If the heading is bare # TLDR, fall back to the first paragraph (1 line, ~80 chars max).

6. Auto-detect tags (2–5)

Scan $TLDR_BODY (case-insensitive) for matches in:

• Languages: elixir, python, javascript, typescript, ruby, go, rust, sql, bash, powershell
• Frameworks: phoenix, ecto, oban, react, next, absinthe, liveview
• Topics: debugging, performance, migration, refactor, investigation, auth, deployment, testing, infrastructure

Pick 2–5 most-mentioned terms. Tags are comma-separated in the next step. sk park-finalize merge-deduplicates these into the manifest entry's existing tags[] array, so any tags accumulated mid-session via /persist survive automatically — the orchestrator does not need to read the manifest or compute the union itself.

7. Finalize via the substrate

sk park-finalize --label "$LABEL" --summary "$SUMMARY" --tags "$TAGS"

sk park-finalize:

1. Reads the ledger (<active-dir>/.session-artifacts.json) as the authoritative artifact list, dedupe-by-name last-write-wins.
2. Verifies each ledger entry exists at <active-dir>/<name> — surfaces missing files as a warning, continues.
3. Renames <sid>-active/ → <YYYY-MM-DD>-<label>/ (with -2, -3, ... collision suffix).
4. Atomic manifest RMW: flips status to archived; populates id, label, summary, date, archive_path, artifacts; merge-dedupes the supplied --tags into the existing tags[] (insertion-order preserved). Respects inherited chain_id, chain_position, previous_session_id, parent_chain_id, checkpoint_nodes. If first-node (chain_position null or 1, chain_id null or == session_id), sets chain_id = <label>.
5. Appends <!-- session-kit-chain ... --> block to ./.stoobz/CONTEXT_FOR_NEXT_SESSION.md (the relay baton stays in cwd for /pickup).
6. Prints summary (Archive path, Artifacts, Relay, Tags, Session UUID prefix, Chain).

Arguments (orchestrator level)

Position	Meaning	Default
<label>	Override the label-resolution chain	(resolved per step 4)

Pass-throughs the orchestrator may set on sk park-finalize:

Flag	When to use
--chain-id <id>	Operator explicitly wants a different chain name (rare)
--no-chain-block	Operator explicitly suppresses chain metadata block (rare)

Exit Codes

sk park-finalize returns:

Code	Meaning	Caller behavior
0	Session archived (or already archived — friendly no-op)	Done
1	Durability failure (rename, mkdir, manifest update)	Surface error; do not claim success
2	Warning (missing ledger artifacts, chain block append failed)	Mention the warning; archive itself is good
3	Usage error	Fix invocation

The three sk write-artifact calls follow their own exit-code table — see write-artifact-protocol.md. If any of them returns 1, abort the orchestrator before calling sk park-finalize — partial archives are not finalizable.

--archive-system — Retroactive Cleanup

This mode is currently unimplemented. When invoked as /park --archive-system, skip the Process above and tell the operator: the flags (--select default, --all, --dry-run, --clean) are reserved with the documented meanings, but no cleanup flow runs. A prior interactive implementation is recoverable from this repo's git history if the mode is needed.

Rules

• All or nothing — Generate all three core artifacts (TLDR + RELAY + HONE). For individual artifacts, use the specific skill directly.
• Always finalize — After writes succeed, sk park-finalize runs automatically. No flag needed.
• No questions — Generate all three without asking. Use best judgment for label and tags.
• Don't re-explain — Just execute. The user wants results, not descriptions.
• CONTEXT_FOR_NEXT_SESSION.md stays in cwd — The mirror is the relay baton /pickup reads next session. sk park-finalize appends the chain block to it after finalization.
• Idempotent re-park — If the session is already archived, sk park-finalize exits 0 with a "nothing to do" message. The original archive is preserved.
• Respect inherited chain metadata — If /pickup inherited chain fields, sk park-finalize keeps them. The substrate handles this; the orchestrator does not need to reason about it.

See also

• tldr/SKILL.md, relay/SKILL.md, hone/SKILL.md — body composition rules
• checkin/SKILL.md — precondition each sk write-artifact triggers internally
• write-artifact-protocol.md — the durable-first contract
• sk park-finalize --help, sk write-artifact --help — full arg / exit-code / JSON reference